Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.
This is not debate club. Unless it’s amusing debate.
For actually-good tech, you want our NotAwfulTech community
I've rather been enjoying following tante's writings lately, and just saw this: https://tante.cc/2024/03/18/5115/
(also I wanted to use the phrase "tech-critical writings" and then didn't because fucking GCs (no, not the memory kind) fucked up the term reference (fucking aaaaarrrrrrggggghhhhh))
@tante is around here at times
yeah I think that's how I found 'em
I haven't seen any numbers on how much Sora cost to train, nor how much it costs to run- I can't help but believe it's painfully high.
last references I saw were that chatgpt et al were $1m/day, and those numbers were pre-sora.
I wish I could cheer for it unambiguously ("yes! burn that money! go ahead! make yourself poorer! woooooo!") but unfortunately the fucking side effects still exist too -_-
Fucking giant water and power guzzling data centers, ugh.
This linked interview of Brian Merchant by Adam Conover is great. I highly recommend watching the whole thing.
For example, here is Adam, decribing the actual reasons why striking writers were concerned about AI, followed by Brian explaining how Sam Altman et al hype up the existential risk they themselves claim to be creating, just so they can sell themselves as the solution. Lots of really edifying stuff in this interview.
Put it this way: The company putting out the product creates an ethics board which warns that the product is too good should ring alarm bells.
I am convinced at this point that Sora is an unfinished mess and the only reason they pushed it out the door is because they needed money.
Didn’t Sammy go grifting in the UAE shortly after?
And then two of their developers alluded that it’s unfinished and said it won’t be released anytime soon, just for their CTO to shut them down two days later and saying, No, no, it’s totally coming later this year.
They could’ve made it do something cool and novel, like a cat riding a pink unicorn fighting dragons. Or probably not, because it can’t do that, so instead they made it their selling point how good it can fake reality during an already juicy election season and applauded themselves for it.
I'm still seeing the same exact issues to a lesser degree in newer models too, which were way more obvious in earlier models. Often by just the "AI smear" (as I like to call it), I can tell AI generated images from the real thing, then I notice the smeared together objects, hands with 6+ fingers (which often are smeared together with other objects), background break, etc.
For me, it tends to be the way stuff is lit. Artists make lighting mistakes too, but the ones AI makes feel more consistent, and different. At least I think that's what usually tipps me off.